An overview of the study design and variables in the GangEpl (Gang Employment) dataset:

This is a long form dataset in which each participant was measured multiple times (indicated by the variable Session). We have 337 observations and 57 variables in total.

Within-subject design: Youth were randomly assigned to either BEP or usual services (US), and data were collected at each intervention session. Hypothesis: Increases in employment among BEP participants would be related to reductions in gang involvement over the course of treatment.

Between-subject design: Compare BEP and US participants at the 3 month and 6 month assessment periods. Hypothesis: BEP would lead to higher employment and less gang involvement than US, and that changes in employment would mediate the effect of treatment on gang involvement.

Variables:

#Updates from Last Meeting

Replicate the graphs in the original paper

The plots I have replicated look roughly the same with the original plots, but they are still not exactly the same.

MLM Analysis with Specified Data Range

The summary of MLM analysis from last time demonstrated that the independent variable EmployHrs (Employment Hours) did not significantly predict the dependent variables Days_Gang (Gang Involvement), Gang_Mem (Gang Membership), and Gang_Activity (Gang Activity). It seems that none of level1 and level2 variables of EmployHrs are significantly predicting the Gang Activity. However, the random effects of level2 variable (Centered Variable of EmployHrs) are significant.

Since the MLM analysis with all participants over all time ranges seems not satisfying (nearly no significant predictions by the IVs), one explanation could be that some participants did not have enough data of time points so their variations were not satisfying for MLM analysis. Pariticipants with ID 10, 13, and 26 were excluded for too few data points. Besides, we excluded session longer than 6 months to further ensure that the dataset had enough variations. Thus, we decided to block off those participants and re-do the analysis. Besides, since this is a longitudinal analysis, we think probably the lad predictors would also contribute to the variation of the DVs. Thus we will include lag predictors in the model.

Gang Involvement

## Warning: Removed 7 rows containing non-finite values
## (stat_smooth).

## Warning: Removed 7 rows containing non-finite values
## (stat_smooth).
## Warning: Removed 7 rows containing missing values
## (geom_point).

## Warning: Removed 7 rows containing missing values
## (geom_point).

Here we can see the overall trend between Employ Hours and Gang Involvement is that for every average increase in employment hours the days with gangs of participants will decrease, although the decreasing trend looks not obvious.

## Warning: Rows containing NAs were excluded from the model.

  • Results of model with covariates Age and Months: It seems that Employment Hours is significant for Gang Involvement at the within-person level, 95%CI [0.00, 0.10].

  • Results of the model with lag predictor EmployHrs_lag1: It looks like the lag predictor of employment hours significantly predicts Gang Involvement, with a 95%CI [0.00, 0.11].

  • Here are some plots of conditional effects of covariates on Gang Involvement.

Gang Activity

## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : pseudoinverse used at 21.171
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : neighborhood radius 17.457
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : reciprocal condition number 0
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : There are other near
## singularities as well. 301.27
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : pseudoinverse used at -0.1068
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : neighborhood radius 5.8846
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : reciprocal condition number 0
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : There are other near
## singularities as well. 17.021
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : pseudoinverse used at -0.10536
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : neighborhood radius 1.2266
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : reciprocal condition number 0
## Warning in simpleLoess(y, x, w, span, degree = degree,
## parametric = parametric, : There are other near
## singularities as well. 402.23

  • The overall trend between Employ Hours and Gang Activity is that for every average increase in employment hours the days with gangs of participants will decrease.
## Warning: Rows containing NAs were excluded from the model.
  • Results of the model with covariate Age and months: It seems that Employment Hours is significant for Gang Activity at the within-person level, 95%CI [0.00, 0.13].
  • Results of the model with lag predictor EmployHrs_lag1: It looks like the lag predictor of Employment Hour does not significantly predict Gang Activity.
## Warning: Predictions are treated as continuous variables in
## 'conditional_effects' by default which is likely invalid for
## ordinal families. Please set 'categorical' to TRUE.

## Warning: Predictions are treated as continuous variables in
## 'conditional_effects' by default which is likely invalid for
## ordinal families. Please set 'categorical' to TRUE.

## Warning: Predictions are treated as continuous variables in
## 'conditional_effects' by default which is likely invalid for
## ordinal families. Please set 'categorical' to TRUE.

## Warning: Predictions are treated as continuous variables in
## 'conditional_effects' by default which is likely invalid for
## ordinal families. Please set 'categorical' to TRUE.

  • Here the point is that there is an obvious decreasing trend in Gang Activity for participants who spend more time in Employment.

Gang Membership

## Warning: Rows containing NAs were excluded from the model.

Results of the model with covariate Age and months: It seems that Employ Hours is significant for Gang Membership at the within-person level, 95%CI [0.00, 0.18].

  • Results of the model with lag predictor EmployHrs_lag1: It looks like the lag predictor of Employment Hour does not significantly predict Gang Activity.

- The conditional effects of Employment Hours on Gang Membership looks really small because the line is nearly close to horizontal.

Short Summary

We have some summary points for MLM analysis with covariates of month, age, and lag predictor of Employment Hours:

  1. It looks like Employment Hours does have a within-person level (centered person-mean0 significant effects on Gang Involvement, Gang Membership, and Gang Activity. The general trends for three paired relationship are both slightly decreasing.

  2. The covariates (age and months) look like not having significant effects on DVs.

  3. The lag predictor of Employment Hours was found significantly predicting Gang Involvement but not Gang Activity and Gang Membership.